fix --nan-equal as in MR30195
(cherry picked from commit f9725df0)
follow up of !62927 (merged), I had not noticed that the original implementation has been subsequently corrected (and the TCT tests in the last nightly indeed failed because of that).
there should have been a similar follow-up MR for 21.0-mc16d but due to a wrong manipulation pushed the commit directly to that branch...